利用層次圖模型進行社群偵測及圖結構之估計

Abstract

圖或者說網絡在社群偵測和圖模型中分別擔任輸入和輸出的角色。由於理解社群結構可提高對圖結構的理解，因此在使用圖模型獲得圖結構之估計後，人們渴望識別潛在的分組。不同於先使用圖模型再對其估計值進行社群偵測，我們的層次圖模型同時估計圖結構和社群結構。該模型將常態-威夏特模型的部分特徵與貝氏社群偵測相融合。最後，我們為後驗推斷開發了一種高效的吉布斯取樣。

Graphs or networks respectively serve as input and output in community detection and graphical models. As understanding community structure enriches our comprehension of graphs, there is a desire to identify potential groupings after obtaining a graph estimate using a graphical model. Rather than sequentially applying a graphical model followed by community structure detection, our hierarchical graphical model concurrently estimates both the graph and community structures. This model blends aspects of the normal-Wishart model with Bayesian community detection. Finally, we develop an efficient Gibbs sampler for posterior inference.